<< output_dims[dnums.output_batch_dimension()]);
  output_dims[dnums.output_feature_dimension()] = kernel_output_feature;
  DCOMMENT("inferrd output output_feature_dimension is "
           << output_dims[dnums.output_feature_dimension()]);

  std::vector<int64_t> input_spatial_dims;
  llvm::SmallVector<int64_t> window_spatial_dims;
  for (auto i = 0; i < num_spatial_dims; ++i) {
    input_spatial_dims.push_back(

    std::iota(broadcast_dims.begin() + axis, broadcast_dims.end(), axis + 1);

    llvm::SmallVector<int64_t, 4> output_dims =
        llvm::to_vector<4>(indices_shape);
    output_dims.insert(output_dims.begin() + axis, depth);

    Location loc = op.getLoc();

    // The iota result is the effective output shape of the computation,

                }
            }
        """

        when:
        succeeds ":app:resolve"

        then:
        def outputDirs = projectOutputDirs("project-artifact.jar", "project-artifact.jar.txt")
        outputDirs.size() == 2
        (outputDirs.parentFile.name as Set).size() == 2
    }

    @ToBeFixedForConfigurationCache(because = "project :lib-project:producer not found.")

  let summary = "Computes gradients for SparseSegmentMean.";

  let description = [{
Returns tensor "output" with same shape as grad, except for dimension 0 whose
value is output_dim0.
  }];

  let arguments = (ins
    Arg<TF_FloatTensor, [{gradient propagated to the SparseSegmentMean op.}]>:$grad,